Vercel supports 30-minute Functions in Fluid compute preview

• Vercel opened a preview that lets Node.js and Python Functions run for up to 30 minutes on Fluid Compute, according to Vercel's announcement post.
• rauchg's thread frames the change as the payoff from a multi-year compute rebuild, with Builds, Sandbox, and now Functions all running on Vercel's microVM-based Fluid infrastructure.
• The new ceiling is aimed at long-running server jobs such as AI workflows, browser automation, OCR, and queue processing, per the official announcement.
• Fluid's billing model still charges for Active CPU only while code is executing, which the official announcement and rauchg's Cloudflare reply both position as the cost story behind longer runs.

You can skim the official announcement, dig into rauchg's architecture thread, and check rauchg's follow-up for the bigger claim that functions, builds, sandboxes, and servers are collapsing onto one compute layer.

The headline is 30-minute Functions, but Vercel spent most of its own explanation on the platform under it. In rauchg's thread, the company says Builds, Sandbox, and Functions now share the same homegrown microVM-based Fluid infrastructure.

That same thread ties the runtime increase to other Fluid features Vercel has already shipped or previewed: multi-concurrency, Active CPU pricing, and Secure Compute for private connectivity. rauchg's follow-up pushes the idea further, arguing that functions, servers, builds, and sandboxes are turning into different configurations of the same underlying compute system.

The mechanics are simple in the docs summary behind the official announcement: durations above 800 seconds are currently in beta, they require Fluid Compute, and the 30-minute limit applies to Node.js and Python.

Configuration stays close to Vercel's existing model:

• Next.js App Router can set export const maxDuration = 1800, according to the official announcement.
• Other frameworks can set maxDuration per path in vercel.json, again per the official announcement.
• The launch examples center on AI reasoning loops, tool-calling, streaming responses, document processing, scraping, and browser automation, per the official announcement.

That makes this feel less like a new product than a wider envelope for the same deployment surface.

In one reply, rauchg's AWS Lambda reply says Vercel pulled this off without using AWS Lambda, while still preserving backward compatibility. In another, rauchg's Cloudflare reply contrasts Fluid with Cloudflare's runtime approach and says Vercel is betting on open runtimes, including bring-your-own runtimes, instead of a pseudo-Node compatibility layer.

The thread context also suggests the ceiling may move again. cramforce's follow-up says the old Fluid limit was 800 seconds, not 15 minutes, and that 30 minutes is not a hard limit.